Extraction of spots in dna microarrays using genetic algorithm

Signal & Image Processing : An International Journal (SIPIJ) Vol.4, No.6, December 2013

EXTRACTION OF SPOTS IN DNA
MICROARRAYS USING GENETIC ALGORITHM
A. Sreedevi1 and Dakshayani. S. Jangamshetti2
1

2

Department of E&E E, R.V. College of Engineering, Bangalore, India
Department of E&E E, Basaveshwar Engineering College, Bagalkot, India

ABSTRACT
DNA microarray technology is an eminent tool for genomic studies. Accurate extraction of spots is a
crucial issue since biological interpretations depend on it. The image analysis starts with the formation of
grid, which is a laborious process requiring human intervention. This paper presents a method for optimal
search of the spots using genetic algorithm without formation of grid. The information of every spot is
extracted by obtaining a pixel belonging to that spot. The method developed selects pixels of high intensity
in the image, thereby spot is recognized. The objective function, which is implemented, helps in identifying
the exact pixel. The algorithm is applied to different sizes of sub images and features of the spots are
obtained. It is found that there is a tradeoff between accuracy in the number of spots identified and time
required for processing the image. Segmentation process is independent of shape, size and location of the
spots. Background estimation is one step process as both foreground and complete spot are realized.
Coding of the proposed algorithm is developed in MATLAB-7 and applied to cDNA microarray images.
This approach provides reliable results for identification of even low intensity spots and elimination of
spurious spots.

KEYWORDS
genetic algorithm, dna microarray, gridding, foreground, background estimation

1. INTRODUCTION
Genomic engineering is an emerging technology which relates biology, medicine and
engineering. Genomic engineering plays a very important role in medical field, such as drug
discovery, gene discovery, diagnosis of disease, effect of drug etc. [1]. The DNA microarray
analysis targets to identify differentially expressed genes in control sample with respect to
reference sample, which can be utilized to study the functions of genes and gene expression
levels. Classical methods deal with analyzing a single gene (probe), whereas DNA microarrays
contain tens of thousands of genes. DNA microarrays are available as single, double and multifluorescent images, depending on labeling of complimentary DNA (cDNA). Double fluorescent
images are most common, in which cDNA are labeled with red (532nm) and green (632nm)
intensities. The cDNA extracted from control and reference samples, labeled with different
fluorescent dyes are hybridized and then spotted on glass slides with robotic means [2].

DOI : 10.5121/sipij.2013.4607

83


In order to arrive at meaningful biological conclusions regarding DNA expression, the initial
steps involved require image processing tools. As the DNA expression is a measure of intensity
of spot, intensities of all the pixels belonging to individual spots have to be measured accurately.
Hence extensive image processing and analysis of microarray images is necessary for reliable
biological interpretations.
Analysis of microarray images comprises of three major steps: gridding, segmentation and
intensity extraction. The techniques for gridding that are most commonly used are only semiautomatic as they require mandatory input parameters such as number of horizontal and vertical
lines and at times manual intervention, in order to locate the grid precisely [3]. Placing the grid is
a challenge as the spots are not uniformly spaced. So most of the techniques are semi-automatic
and may involve human intervention for choosing the position for the grid, since gridding forms
unevenly spaced parallel and perpendicular lines. Also the non-uniformity in the position of the
spots makes it almost impossible to get a grid, such that a single complete spot is positioned in an
individual grid cell. Gridding is followed by spot segmentation techniques. ScanAlyze software
uses a fixed circle segmentation method while GenePix uses adaptive circle segmentation
method. Both the methods are not idyllic for measuring spot intensities of non circular spots.
Gene expression data derived from arrays may be used for gene clustering, cancer detection and
other analysis. Then DNA microarrays provide a medium for matching known and unknown
DNA samples based on base-pairing rules and automating the process of identifying the
unknowns.

2. LITERATURE SURVEY
During the microarray slide preparation non-uniformity in spot positioning is caused by the
equipment, since the pins in the spotting machine might bend over the time leading to irregularity
in spacing between the spots. Under such conditions template based approaches for grid
formation may lead to inaccurate results [4-6]. The limitation of microarray image analysis is due
to cumulative errors from numerous sources; hence often require manual analysis of array image
data to ensure accuracy [7].
There is no direct approach for both gridding and spot segmentation. It is difficult to obtain
optimal location of the grid location due to non uniformity in spot location. Hence it is a
complicated, time consuming process and requires human intervention.
Automatic gridding techniques are available like hill climbing approach [8], and gridding using
genetic algorithms (GA) [9]. Since last decade, GA is implemented to search for optimum value
in a given search space. GA is more suitable when the search space is large because of its parallel
searching capability. GA search gives global optimal solution. Genetic algorithms are stochastic,
robust optimizers, suitable for solving problems, where there is little or no prior knowledge is
available.
In our earlier work, an approach based on GA consists, a single step wherein the spots are located
without the formation of gridding [10]. This paper presents a method for identification of
foreground and background pixels of spots and spot centers along with its mean values.

84


3. GENETIC ALGORITHM AND ITS IMPLEMENTATION
The work concentrates on detailed way in which using GA, spots in the microarray are identified
and the estimation of foreground and background has been computed.
The following steps are involved in Image Analysis of DNA microarray images
1. Preprocessing
2. Identification of spots
3. Segmentation
4. Finding the center and average intensity
5. Removal of spurious spots
6. Background estimation

3.1 Preprocessing
An ideal spot representing DNA should contain pixels of uniform intensity. But hybridization and
slide preparation leads to non uniform intensities of the pixels belonging to an individual spot. So
preprocessing is necessary to remove non uniformity and the noise.
Microarray image is preprocessed with Median filter which is an order stochastic filter and non
linear in nature. Median filter replaces the value of a pixel by the median of gray levels in the
neighborhood of that pixel. Median filters provide excellent noise reduction capability for random
and salt and pepper noises with considerably less blurring than linear smoothing filters of same
size. Hence edges are preserved and noise is eliminated. Also isolated clusters of pixels with an
area less than n2/2 are eliminated by an n x n median filter, hence even spurious spots can be
removed to a larger extent.

3.2. Identification of spots
Identification of spots is carried out using GA approach without formation of grids. Genetic
algorithm searches fitness function which is the combination of objective function and constraint
function. Finding a pixel belonging to a spot is a maximization problem in GA. A part of DNA
Microarray image of size n x m pixels is considered. Certain number of pixels having total fitness
greater than the threshold value is encoded as a chromosome.
3.2.1. Chromosome:
The "chromosomes" encode a group of linked features. In our work chromosome is encoded as an
array of binary digits. Coding consists of number of strings, each string representing either row or
column of a pixel. Number of strings is decided by the number of pixels considered for a
chromosome.
Figure 1 shows sample coding when two pixels are considered for a chromosome. Row and
column numbers of each pixel is coded in binary. Number of bits of each string and hence length
of chromosome depends on the size of the sub image chosen for processing.

85


For example a sub image of size 31 x 31 and two pixels per chromosome are considered for
processing.
Length of each substring = 5 bits
Length of each string (chromosome) = 20 bits

Figure1. Coding of one chromosome

3.2.2. Population:
Set of chromosomes form the population. Information of set of pixels in a sub image is taken as a
chromosome. Initial population is randomly selected. Three genetic operators with elitism
technique are realized. Genetic operators Figure 2 gives flow chart of procedure to obtain spot
using GA.
Selecting number of copies of an existing chromosome is done by Roulette Wheel selection
method to obtain parent population for the purpose of crossover in order to obtain next generation
of chromosomes with better fitness. This process is repeated till the maximum number of
iterations, giving the result as a pixel within a spot. Genetic operators like mutation and crossover
are applied to get best results. Number of chromosomes in a population is maintained constant for
all the generations.
After the crossover is performed, mutation takes place, which prevents falling all solutions in
population into a local optimum. Mutation changes the new offspring randomly. In case of binary
encoding, few randomly chosen bits are switched from 1 to 0 or from 0 to 1.

86


Figure 2. Flowchart to identify spots (step2)

3.2.3 Objective function:
Objective function is combination of fitness function and constraint.
Fitness function:
The fitness f(i) of a particular chromosome which is derived for the solution of optimization
problem, is defined by the equation (1)

87


GA searches for best value of objective function Objective function with violated row and
column values will be penalized to discourage the solution. Genetic Algorithm searches for
optimal locations of a pixel belonging to a spot by maximizing the fitness function.
The values assigned for variables used in the GA method are given below. The termination of the
program is decided by maximum number of generations and the size of the population gives the
information of number of search points. The application of genetic operators is dependent on the
respective probabilities. Scaling is used to obtain appropriate fitness value.
Maximum number of generations
Maximum population size
Crossover probability
Mutation probability
Scaling

=
=
=
=
=

60 to 100
40
0.08
0.03
1.1

3.2.4. Segmentation:
Individual spot is identified by a pixel, which is obtained by applying method developed, as
described in step2. Once the complete image is processed by the GA method, all the spots are
identified.
Segmentation process extracts the pixels representing the spot. The pixels belonging to a spot are
considered as foreground. All these pixels in a spot have the same intensity in an ideal case.
To identify the foreground pixels of a spot, a specific local threshold value is selected depending
on the intensity level of spot. Both minimum and maximum values of intensity are derived to
88


distinguish between neighboring spots. The algorithm scans for the pixels lying between two
threshold values and distinguishes between foreground and background. Once all the foreground
pixels are identified mean or median of spot intensity is calculated.
Threshold values are calculated by following equations
Minimum = inty_p - 0.4*inty_p
Maximum = inty_p + 0.2*inty_p
Where inty_p is the intensity of the pixel obtained from GA.
3.2.5. To find the center and average intensity:
Mean intensity of each spot is calculated from the output of foreground estimation. Centre is
determined from extreme rows and columns. Average spot intensities along with their respective
centre co-ordinates are shown in figures 4, 5 and 6 for the different sizes of sub-images
considered.
3.2.6. Removal of spurious spots:
The spots with very low intensity and small size are considered as spurious spots. The spurious
spots are eliminated and the actual spots are considered.
3.2.7. Background estimation:
Intensity of the spots depends on signals from scanner devices originating due to fluorescent
molecules attached to hybridize DNA, signals due to coating of glass and contamination in
hybridization and washing processes. Even limitation of scanner bandwidth plays an important
role in background estimation [12]. Hence measured spot intensity is true intensity of the spot
plus its local background.
Basically there are two methods to estimate the background. In one of method histogram is used,
in which range of pixel intensities in the image is obtained. Hence global background can be
calculated. Where as in the other, pixels close to spot are assumed to be the local background. In
the process of estimation of background, normally local background level is assumed to be same
as that of the intensity of the pixels in the proximity of the corresponding spot.
In this approach the background of each spot is obtained from the foreground pixel data and the
complete spot. The mean (or median) intensities for foreground and background are calculated.
The actual intensity of the spot is the difference in foreground and background mean (or median)
values.

4. RESULTS
In this paper a part of microarray image is considered for which the results are evaluated and
presented. Figure 3 shows the part of DNA microarray image of size 200x90.
GA algorithm is implemented on sub image of size n x n to identify the spots. The process is
repeated for entire image to extract all the spots. This algorithm is applied to images of 3 different
89


sizes of the sub images and the results are discussed. Figures 4, 5 and 6 show the average
intensity of spots plotted at their respective centre co-ordinates, for sub image of size 31x31,
15x15 and 7x7 respectively.

Figure 3. Part of microarray image
(Original image courtesy of Prof. Paturu Kondaiah,
Dept of M.R.D.G, IISc, Bangalore)

Figure .4. Average intensities of spots for sub image of size 31 x 31

90


Figure 5. Average intensities of spots for sub image of size 15 x 15

Figure 6. Average intensities of spots for sub image of size 7 x 7

Figure 7 shows the complete spot extracted by knowing the pixel features belonging to that spot
from G.A. Once the spot is identified, foreground of the spot is estimated by implementing
automatic segmentation technique. Figure 8 shows the foreground and figure 9 shows the
corresponding background.

91


Figure 7. Complete spot

Figure 8. Foreground

Figure 9. Background

5. CONCLUSION
GA with elitism technique is implemented on the cDNA Microarray image. Elitism removes
pixels with low fitness from the off springs and is replaced by pixels with larger fitness (parents)
to obtain next generation. This technique results in extraction of spots accurately. From the results
it is evident that sub image of size 7x7 identifies all the spots but requires more time for
processing. Appreciable results are obtained even with size 15x15. When sub images of size
31x31 are processed, only certain spots are realized as shown in Figure.4. Figures 7, 8 and 9 show
one complete spot, its foreground and background respectively. The local background gives
reliable results compared with global background.

REFERENCES
[1]

[2]
[3]
[4]
[5]
[6]

[7]
[8]

[9]

Xin-Yuzhang, Fei Chen, Yuan-Ting Zhang, Shannon C. Agner, Metin Akay, Zu-Hong Lu, Mary Miu
Yee Waye, and Stephen Kwok-Wing Tsui, “Signal Processing Techniques in Genomic Engineering”,
Proceedings of IEEE, 0018-9219/02©2002 IEEE
Mohammed Khabzaoui, Clarisse Dhanens, and El-Ghazali Talbi, “A Multicriteria Algorithm to
analyze DNA Microarray data”, 0-7803-8515-2/04©2004 IEEE
Peter Bajcsy, “Automatic grid alignment in DNA Microarray Scans”, IEEE Transactions on image
processing, Vol 13, 1057-7149/04©2004 IEEE.
Buhler J., T. Ideker , and D. Haynor, “dapple: Improved Techniques for finding spots on DNA
Microarrays”, UV CSE Technical Report UWTR 2000-08-05.
Scanalytics Inc, “Microarray Suite”, Product description at
http://www.scanalytics.com/product/hts/microarray. html
Peter Bajcsy, “An Over view of DNA Microarray Image requirements for Automated Processing”,
Proceedings of the 2005 IEEE computer society conference on Computer vision and pattern
Recognition.
Christian Uehara and Ioannis Kakadiaris, “Towards Automatic Analysis of DNA Microarrays”,
Proceedings of Sixth IEEE workshop on Applications of Computer Vision 2002.
Luis Rueda and Vidya Vidyardharan, “A Hill Climbing Approach For Automatic Gridding Of cDNA
Microarray Images”, IEEE/ACM transactions on computational Biology and Bioinformatics, Vol.3,
No.1, January-March 2006.
E.Zacharia and D. Maroulis, ”An Unsupervised and Fully- Automated image Analysis method for
cDNA microarrays”, Twentieth IEEE International Symposium on Computer Based Medical Systems
2007.

92

[10] A.Sreedevi and D.S.Jangamashetti, “Automatically locating Spots in DNA microarray Image Using
Genetic Algorithm without Gridding”, IACSIT spring conference, 978-0-7695-3653-8/09 © 2009
IEEE .
[11] Anders Bengtsson and Henrik Bengtsson, “Microarray image analysis: background estimation using
quantile and morphological filters”, BMC Bioinformatics 2006, 7:96
AUTHORS
Dr. A.Sreedevi, born in Andhra Pradesh, India in 1961. She received her Electrical
Engineering degree from Bangalore University during 1983, Her M.E degree from Pune
University during1993 and Ph.D. from Visveswaraya Technological University, India.
Since 1984 she is in teaching, presently working as Associate professor in the
Department of Electrical and Electronics Engineering, RV College of Engineering,
Bangalore. Her ares of research are signal and image processing.
Dr. D.S. Jangamshetti, born in Karnataka, India in 1964. She received Electrical
Engineering degree from Karnatak University during 1985, M.Tech. degree from I.I.T.
Khargpur, and Ph.D. from I.I.T. Bombay in 2003. She is presently working as professor
in the Department of Electrical and Electronics Engineering, Basaveshwar Engineering
Collage, Bagalkot, Karnataka.
She has several publications in IEEE Transactions. Her area of interest is signal
processing. She has attended a number of international and national conferences. She has organized quite a
few short term training programs for teachers and students. She is senior member of IEEE, member of
B.M.E., I.S.T.E. and Institution of Engineers,India.

93

Extraction of spots in dna microarrays using genetic algorithm

More Related Content

What's hot (19)

Viewers also liked (19)

Similar to Extraction of spots in dna microarrays using genetic algorithm (20)

Recently uploaded (20)

Extraction of spots in dna microarrays using genetic algorithm